CellBRF: a feature selection method for single-cell clustering using cell balance and random forest.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed analysis of biological tissues.
• Accurate cell sub-population identification relies on effective feature selection for clustering.
• Current methods often fail to fully leverage gene discriminatory power across cell types.

Purpose of the Study:

• To develop a novel feature selection method, CellBRF, for improved single-cell clustering.
• To enhance the accuracy and interpretability of scRNA-seq data analysis.
• To address limitations in existing feature selection approaches for cell type identification.

Main Methods:

• CellBRF utilizes random forests to identify genes crucial for discriminating cell types.
• Incorporates a class balancing strategy to handle unbalanced cell type distributions.
• Evaluated on 33 diverse scRNA-seq datasets.

Main Results:

• CellBRF significantly outperforms state-of-the-art feature selection methods.
• Demonstrates superior clustering accuracy and cell neighborhood consistency.
• Successfully applied in case studies for identifying cell differentiation stages, non-malignant subtypes, and rare cells.

Conclusions:

• CellBRF offers a new, effective approach to boost single-cell clustering accuracy.
• The method provides valuable selected features for various scRNA-seq analysis tasks.
• CellBRF is freely available, promoting its adoption in the research community.